Surge Classification

After talking with Mike last week I decided to go back to my surge classification work. Along with thresholds for dewpoint and winds, I have now included precipitable water. Before I get to how I have taken into account precipitable water, I first want to mention how I have come to my threshold values for dewpoint and winds.

Many of the Stensrud papers have used wx station data from Yuma, AZ to define surges. Since this is the present benchmark for surge classification, I have worked to judge my results to theirs, i.e., roughly the same average monthly surge events. All of the papers that I have read so far have found an average of ~3.0 surge events per month for July and August (Note: Almost all of these papers only look at data for these two months). Mike, though admitting that describing an average for something as dynamic and variable as the NAM can be tricky, said that this value seemed quite legitimate. As a simple test of the methods of Stensrud, I ran his threshold algorithm on our SAO data for the Yuma station and got very similar results. After this, I used data from the closest NARR grid point to Yuma. With some trial and error I found that a 70th percentile threshold level for dewpoint, wind speeds and precipitable water gave me comparable results. From this successful result I moved forward and defined a larger domain of interest to test these thresholds over a larger, and more appropriate, area.

I am using a domain of 31 to 35N and 108 to 115W. An example of this domain is shown below for precipitable water, in mm, on 19 August 1983.
I calculated an aggregated daily mean for each variable, as well as a daily 70th percentile value for each variable. Events of each are defined when the daily value is equal to or greater than the 70th threshold. I defined 4 different combinations of these criteria:

(1) Wind event on Day 1; PW event on Day1 or Day 2; DWP event on Day 1
(2) Wind event on Day 1; PW event on Day1 or Day 2; 2 consecutive days of DWP event from Day 1
(3) Wind event on Day 1; PW event on Day1 or Day 2; 3 consecutive days of DWP event from Day 1
(4) Wind event on Day 1; PW event on Day1 or Day 2; DWP event on Day 1 or Day 2

The average monthly number of surge events in a given July or August for each criteria were 2.85 (1), 2.79 (2), 2.2 (3) and 3.86 (4). Next, I wanted to compare these results with the results from using the Stensrud surge criteria for the NARR grid point near Yuma. I did this by finding days were an event was classified by a specific criteria about, i.e., 1 through 4, and also classified by the one grid point Stensrud criteria for Day-1, Day or Day+1. The results for percentage of agreement is below for each criteria:

(1) 59.8%
(2) 64.7%
(3) 70.7%
(4) 56.9%

There does appear to be rather good agreement between these criteria and Stensrud's results, but it is not perfect. The criteria with the lowest average monthly surge events, (3) with 2.2, showed the highest correlation with Yuma. In general, I believe there is much room for improvement over the classification scheme of Stensrud since he has used such a limited data source of only one station. Branching out from what is tried and true could raise a lot of questions of our work, but I imagine that is typical of anything new within research.